Word Recognition

1
Word Recognition

• Rauno Parrila
• University of Tromsø
• raunop_at_psyk.uit.no

2
Contents

• 1. Definitions
• 2. Letter recognition
• 3. Sublexical units in word recognition
• 4. Lexical factors
• 5. Semantic factors
• 6. Language level factors

3
Why word recognition?

• Word recognition literature cuts across several
  basic research areas in cognitive psychology
• verbal learning and memory
• memory access (via phonemes, morphemes,
  graphemes, semantics etc.)
• attention (automatic vs. controlled processing)
• pattern recognition (features vs. templates)

4
Definitions

• Phone
• elementary speech sound or sound unit (acoustic
  unit)
• Phoneme
• a group of speech sounds spelled with the same or
  equivalent letter and commonly regarded as the
  same sound. They may vary somewhat (be different
  phones) but do not differentiate between
  meanings.
• Allophones
• all phones not distinguished in a language as
  separate phonemes

5

• Definitions cont.
• Homophones
• words that sound the same but are spelled
  differently
• Pseudohomophone
• nonword that sounds like a real word when
  pronounced
• Onset
• initial consonant or consonant cluster in a
  syllable
• Rime
• everything that follows in a syllable after onset

6

• Definitions cont.
• Phonetics
• study of raw sounds (phones)
• Phonology
• study of sounds within a language (phonemes)
• Grapheme
• a letter or combination of letters that represent
  a phoneme (basic unit of written language)
• Morpheme
• unit of structure that reflects meaning

7

• Definitions cont.
• Morphology
• study of words and word formation
• Semantics
• study of meaning
• Pragmatics
• study of language use
• Syntax
• study of word order

8
Word Recognition The Task

• 1. Recognize input stimuli as letters (pattern
  matching)
• 2. Recognize combination of letters as a word
  (visually or via phononological recoding)
• 3. Retrieve the meaning from lexicon (lexical
  access)
• 4. Retrieve phonological representation (sound
  lexicon)
• 5. Assemble motor program for pronunciation
• 6. Execute pronunciation program

9
WRITTEN WORD
Naming
10
Lexical Decision
11
Visual Analysis SystemLetter Recognition

• How do we recognize a group of lines and curves
  as letters?
• Two common explanations Template matching and
  feature detection
• Reisberg, , Chapter 2

12
Template matching

• Stored representation in brain for every letter
  (every version of that letter)
• Costly think of all the possible fonts,
  handwriting styles etc.
• Normalization before matching
• Perhaps enough space for letters but for all
  visual stimuli? Two different systems for letters
  and other visual stimuli?

13
(No Transcript)
14
(No Transcript)
15
Feature detection

• analysis-by-synthesis
• 1. Letter broken down to its constituent parts
• 2. List of parts compared to patterns in memory
• 3. Best matching pattern chosen

16
Cognitive demons shout when they receive
certain combinations of features
Feature demons decode specific features
Image Demon receives sensory input and sends
signal further
Decision demon listens for the loudest shout in
pandemonium to identify input
17
A
18
(No Transcript)
19
Word recognition Sounds

• Grapheme-phoneme correspondence
• print to sound conversion by rules or analogies
• Spelling-sound regularity effect
• words with consistent spelling-sound relations
  are read aloud faster (by skilled readers) than
  irregular words
• no difference in lexical decision?
• Pseudohomophone effect
• Lexical decision speed slower with nonwords that
  sound like real words than with real words or
  with nonwords that do not sound like real words
• Homophone categorization

20
Word recognition Groups of letters

• word superiority effect with words and
  orthographically regular (pronouncable) nonwords,
  but not with unpronouncable nonwords (Eysenck
  Keane, p. 292)
• non-word legality effect (orthographic regularity
  effect)
• response time for nonwords following spelling
  patterns of real words longer than to nonwords
  that are not word-like (non-word legality effect)
  
• perhaps orthographic processing effect in
  general words containing frequent letter bigrams
  or trigrams recognized faster than other words
• positional frequency effect

21
HIGHER LEVEL INPUT
McClelland Rumelhart, 1981
WORD LEVEL
PHONEME LEVEL
LETTER LEVEL
ACOUSTIC FEATURE LEVEL
FEATURE LEVEL
ACOUSTIC INPUT
VISUAL INPUT
22
BAD
CAB
A
D
B
C
E
23
Do other sublexical units help in recognizing
printed words?

• Sublexical units bigger than phonemes and
  graphemes?
• onsets and rimes
• onset initial consonant or consonant cluster in
  a word or syllable
• rime following vowel and consonants
• if words broken at onset-rime boundary, resulting
  letter clusters more easily recognized as
  belonging together than if broken at other points
• example FL OST ANK TR
• vs. FLA ST NK TRO

24

• syllables no role in visual word recognition?
• morphemes
• root morpheme effect in lexical decision
• morphemic priming effect
• JUMP and JUMPED both prime JUMP

25
(No Transcript)
26
Lexical Level Variables

• Word frequency effect
• naming latency for high-frequency words shorter
  than for low-frequency words
• lexical decision time faster for high-frequency
  words than for low-frequency words
• shorter fixation durations for high-frequency
  words than for low-frequency words when reading a
  passage
• Word familiarity effect
• familiarity ratings usually based on subjective
  ratings (spoken word frequency) and frequency
  ratings to analysis of texts (written word
  frequency)
• independent familiarity effects in naming and
  lexical decision tasks (after frequency
  controlled)

27
(No Transcript)
28
ENG ENG
29
(No Transcript)
30
ASC ASB
31
(No Transcript)
32
ONG ONG
33
(No Transcript)
34
RAS RAD
35
(No Transcript)
36

• Lexical status effect
• response time to words faster than to nonwords in
  lexical decision tasks
• words named faster than nonwords (even
  pseudo-homophones)

37
Why lexical access faster for high frequency and
familiar words?

• activation explanation
• high-frequency and familiar words will have
  higher resting level activations due to increased
  experience with them
• ordered search (list models)
• lexicon searched serially with high-frequency and
  familiar words being searched before
  low-frequency and unfamiliar words

38
LION
Orthographic access file
Phonological access file
Semantic access file
LION
/l/i/o/n/
lion
39

• Semantic variables
• Neighbourhood effect (word similarity effect)
• nonwords with large neighbourhood take longer
  to reject than nonwords with small neighbourhoods
• low frequency words with large neighbourhood
  are recognized faster than low frequency words
  with small neighbourhoods
• Ambiguity (meaningfulness) effect
• more meanings, faster recognition
• Repetition priming
• familiar word encountered for the second or third
  time in a task is named and recognized faster
  than when it was encountered the first time

40

• Semantic (context) priming
• words appearing at the end of constraining
  sentence are recognized faster than otherwise
• eye fixations shorter (or skipped) when words
  highly compatible with context
• irrelevant context makes target words harder to
  recognize
• words appearing in isolation recognized and named
  faster if preceeded by a semantic associate
• difficult to explain for the ordered search models

41
Summary

• Two routes to word recognition (lexicon)
• 1. Direct visual access
• Lexical route, direct route, recognition by sight
  (sight-words), visual route
• Assumption is that some number of (common) words
  are recognized as visual units without access to
  sound based information
• 2. Grapheme-phoneme conversion
• print-to-sound conversion, spelling to sound
  conversion, speech recoding, phonological
  recoding, phonological coding, deep phonological
  coding, phonetic recoding
• lexicon accessed through speech based code

42
Which route is chosen?

• Differences between words (stimuli)
• Differences between individuals
• Differences between languages Orthographic depth
  of the language
• 1. shallow orthography the phonemes of the
  spoken word are represented by the graphemes in a
  direct and unequivocal manner (isomorphism)
• 2. deep orthography spelling to sound relations
  (more) opaque (e.g., same letter may represent
  different sounds in different contexts, or
  different letters may represent the same sound in
  different contexts)

43

• Frost, Katz, Bentin (1987)
• Compared visual word recognition in
  Serbo-Croatian (shallow orthography), Hebrew
  (deep orthography), and English (medium)
• found that word frequency effect (high vs. low)
  and lexical status effect (words vs. nonwords) on
  naming speed depended on language largest on
  Hebrew, smallest on Serbo-Croatian
• performance in lexical decision and naming tasks
  almost identical in Hebrew and clearly different
  for Serbo-Croatian
• semantic priming effected naming speed in Hebrew
  and English but not in Serbo-Croatian

44
WRITTEN WORD
Serbo-Croatian (English)
VISUAL ANALYSIS SYSTEM
VISUAL INPUT LEXICON
SEMANTIC SYSTEM
GRAPHEME-PHONEME CONVERSION
SPEECH OUTPUT LEXICON
PHONEME LEVEL
SPEECH (NAMING)
45
WRITTEN WORD
Hebrew (English)
VISUAL ANALYSIS SYSTEM
VISUAL INPUT LEXICON
SEMANTIC SYSTEM
GRAPHEME-PHONEME CONVERSION
SPEECH OUTPUT LEXICON
PHONEME LEVEL
SPEECH (NAMING)